Starling LM 7B Alpha

About

Starling-LM-7B-alpha is an advanced large language model developed by Berkeley NEST, utilizing the Openchat 3.5 framework, which builds on Mistral-7B-v0.1. It incorporates Reinforcement Learning from AI Feedback (RLAIF) and is fine-tuned with the Nectar dataset, comprising 183,000 chat prompts and 3.8 million pairwise comparisons. It excels in conversational tasks, scoring 8.09 on the MT Bench benchmark, surpassing most models except OpenAI's GPT-4 and GPT-4 Turbo. Despite its prowess in dialogue, content generation, and question answering, it struggles with reasoning and mathematics and sometimes produces verbose responses or is vulnerable to jailbreaking prompts, highlighting areas for potential enhancement 1 2 3.

Capabilities